Predictive and prognostic role of tumour-infiltrating lymphocytes in breast cancer patients with different molecular subtypes: a meta-analysis

Background Whether tumour-infiltrating lymphocytes (TILs) play different roles in different molecular subtypes of breast cancer remains unknown. Additionally, their prognostic and predictive value in different molecular subtypes of breast cancer is still controversial. The aim of our meta-analysis was to assess the prognostic and predictive value of TILs in different molecular subtypes of breast cancer by summarizing all relevant studies performing multivariate analysis. Methods PubMed, Embase, EBSCO, ScienceDirect, the Cochrane Database and Web of Science were comprehensively searched (until March 2020). Hazard ratios (HRs), odds ratios (ORs) and their 95% confidence intervals (CIs) were used as effect measures to perform our meta-analysis. A random effect model was used. Stata software, version 15 (2017) (StataCorp, College Station, TX, USA) was used to perform the statistical analysis. Results Thirty-three studies including 18,170 eligible breast cancer patients were analysed. The meta-analysis showed that high TIL expression was significantly associated with increased pathological complete response (pCR) rates after neoadjuvant chemotherapy in patients with the HER2-enriched molecular subtype (OR = 1.137, 95% CI [1.061 ~ 1.218], p < 0.001) and triple-negative breast cancer (TNBC) subtype (OR = 1.120, 95% CI [1.061 ~ 1.182], p < 0.001). However, high TIL expression was not significantly associated with high pCR rates after neoadjuvant chemotherapy in patients with the luminal molecular subtype of breast cancer (OR = 1.154, 95% CI [0.789 ~ 1.690], p = 0.460). We carried out a meta-analysis on the HRs of overall survival (OS) and disease-free survival (DFS) to assess the prognostic value of TILs in breast cancer with different molecular subtypes more deeply. Our meta-analysis confirmed that high TILs were associated with significantly improved DFS in patients with the HER2-enriched molecular subtype [HR = 0.940, 95% CI (0.903 ~ 0.979), p = 0.003] and TNBC molecular subtype [HR = 0.907, 95% CI (0.862 ~ 0.954), p < 0.001]. However, high TILs were not associated with significantly better DFS in patients with the luminal molecular subtype of breast cancer [HR = 0.998, 95% CI (0.977 ~ 1.019), p = 0.840]. Furthermore, the results confirmed that high TILs were significantly related to better OS in patients with the HER2-enriched molecular subtype [HR = 0.910, 95% CI (0.866 ~ 0.957), p < 0.001] and TNBC molecular subtype [HR = 0.869, 95% CI (0.836 ~ 0.904), p < 0.001]. Conversely, the summarized results indicated that high TILs were significantly associated with poor OS in patients with the luminal molecular subtype of breast cancer [HR = 1.077, 95% CI (1.016 ~ 1.141), p = 0.012]. Conclusions Our meta-analysis confirms that high TILs are associated with favourable survival and predicts pCR in breast cancer patients with the TNBC and HER2-enriched molecular subtypes.

Keywords: Breast cancer, Tumour-infiltrating lymphocytes, Molecular subtype, Prognosis, Prediction, Meta-analysis Background Breast cancer is one of the most common malignant tumours in women [1] and is still the second leading cause of cancer-related death in women around the world [2]. At present, the forecast of prognosis is not ideal, and a specific predictor is needed to enhance the individualized therapeutic effect. The complex interaction between the immune system and cancer cells plays a vital role in controlling and eradicating cancer and is regulated by a delicate balance between activation and suppression signals [3]. Research on the microenvironment of tumours can reveal the complex correlation between the immune system and the biological behaviour of cancer cells. To restrict the development of breast cancer, it is very important to understand the tumour microenvironment.
Increasing evidence indicates that the tumour microenvironment plays an important role in tumour formation, growth, invasion and metastasis. Tumourinfiltrating lymphocytes (TILs) have emerged as potentially important prognostic and/or predictive biomarkers for breast cancer [4,5]. Although valuable information has been obtained, the heterogeneity in experimental design and TIL assessment has hindered a more comprehensive understanding of the biological value of TILs. However, the prognostic value of TIL remains complex and controversial. Breast cancer is a clinically and molecularly heterogeneous disease, and various factors determine the prognosis and response to treatment.
Thus, we carried out this meta-analysis, aiming to estimate the prognostic and predictive value of TILs in patients with different molecular subtypes of breast cancer.

Retrieval strategy
Embase, PubMed, EBSCO, the Cochrane Database, ScienceDirect and Web of Science were comprehensively searched for studies exploring the prognostic and predictive relationship between TILs and the different subtypes of breast cancer (without time, place of publication or language restrictions; until March 2020). No retrieval restrictions were used. In addition, the reference lists of searched reviews and studies were examined to further identify potentially related articles. The main retrieval terms applied were "breast cancer" or "breast carcinoma" and "neoadjuvant chemotherapy" and "TILs" or "Tumorinfiltrating lymphocytes" and "prognosis" or "change".

Selection standards
To ensure the accuracy and reliability of our analysis, we selected qualified studies based on the following criteria. (i) The prognostic or predictive value of TIL testing in different subtypes of breast cancer with at least one relevant outcome indicator was reported in the research or could be computed based on published data. (ii) The studies were of high quality and performed multivariate analysis on pathological complete response (pCR) or survival data such as disease-free survival (DFS) or overall survival (OS). (iii) The hazard ratio (HR), odds ratio (OR) and their 95% confidence intervals (CIs) were reported or could be calculated according to the outcome data (DFS, OS or pCR). (iv) The samples were taken from core-needle biopsy specimens or surgical specimens after the operation.
Two authors (Zhao-hua Gao and Ming Liu) independently performed the literature retrieval and determined qualified studies according to the inclusion criteria. Any disagreements between the authors were settled by discussion and consensus. If no agreement could be reached, the final outcome was determined by a thirdparty researcher (Cun-xin Li). If there was more than one publication on the basis of the same patient groups, the most informational research was used.

Research quality appraisal and data collection
The data were collected according to the Cochrane guidelines [6]. Two authors (Zhao-hua Gao and Ming Liu) examined the eligible studies independently, and any disagreements between the authors were settled by discussion and consensus. The following data were collected for our meta-analysis: publication time, first author, country, study design, baseline patient characteristics, age range, treatment type, molecular subtypes, ethnicity, follow-up duration, TIL cut-off value, TIL position, outcomes (pCR, DFS, or OS), HR, OR and 95% CI. The Newcastle-Ottawa scale (NOS) criteria were used to evaluate the quality of the selected eligible studies [7]. A funnel plot was used to estimate the publication bias. The studies selected in our meta-analysis obtained written informed consent from all patients and were carried out according to clinical practice principles, all local regulations and the Helsinki Declaration.

Statistical analysis
In this meta-analysis, we chose pCR as a predictor of neoadjuvant chemotherapy (NAC) for breast cancer. We assessed the overall OR and its 95% CI of the qualified studies to analyse the predictive value of TILs for NAC in breast cancer. OS and DFS were used as prognostic outcomes in our meta-analysis. In the meta-analysis, the HR and its 95% CI were used as the effect scales of prognosis. The associations between TILs and clinicopathological parameters were evaluated using ORs and 95% CIs. If the HR or OR and its 95% CI could not be obtained directly from the original article, we used the available data to calculate them with the software designed by Tierney et al. [8]. The Q test was used to estimate the heterogeneity between studies, and the I 2 value represents the size of the heterogeneity [9]. I 2 values > 40% indicated high heterogeneity [6]. If the heterogeneity was high, a random effect model was used; if not, a fixed effect model was used. The P value was set as < 0.05 to indicate statistical significance. The clinicopathological parameters and predictive and prognostic indicators of all relevant studies were pooled and analysed. At the same time, subgroup analysis was completed based on different countries and different study designs. The quality and homogeneity of the results were assessed by sensitivity analysis. A funnel plot was used to test publication bias. In the statistical analysis, we referred to the statistical parameters and methods used by our team in previous studies [10].
Stata software, version 15 (2017) (StataCorp, College Station, TX, USA) was used to carry out the statistical analysis. This meta-analysis followed the recommendations of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines [11].

Baseline characteristics of the qualified studies
In the systematic literature retrieval, we found 617 studies. By reviewing the titles and abstracts, 74 possible related studies were identified. Of these 74 studies, 41 studies were later excluded because they did not meet the selection criteria. Eventually, we determined that 33 studies met the inclusion criteria [4,5,. Fig. 1 summarizes the search and screening process. The 33 studies comprised 18,170 qualified patients with breast cancer (sample capacity, median: 331 [50-3771], mean: 550). These studies were published between 2010 and 2020 and were from Europe, America, Australia and Asia (Spain, France, Italy, United Kingdom, Belgium, Finland, Germany, Ireland, USA, Canada, Australia, Japan, Korea, and China). The eligible studies evaluated TILs by haematoxylin and eosin-stained sections. Twelve studies provided ORs for pCR to complete the meta-analysis [14, 24, 26, 27, 29, 32, 34-36, 38, 39, 42]. Fifteen of these studies provided HR data for DFS or OS, and we performed the pooled analysis. Twelve studies provided HR data for DFS [5, 14, 16-18, 25, 27, 30-32, 38, 41], and ten studies provided HR data for OS [5,14,16,18,19,21,22,25,27,30]. Table 1 summarizes the main baseline characteristics. We evaluated the quality of the selected studies based on the NOS, as shown in Table 2.

Relationship of lymphocyte-predominant breast cancer (LPBC) with clinicopathological parameters T stage
The incidence of LPBC in the T3 and T4 groups was lower than that in the T1 and T2 groups, and the difference was statistically significant (OR = 0.646, 95% CI In the Asia and Europe groups, the differences were statistically significant.

Lymph node status
The pooled analysis indicated that the incidence of LPBC detection between the lymph node metastasis group and the non-lymph node metastasis group was not significantly different (overall: OR = 0.941, 95% CI The difference was statistically significant in the America group.

Impact of TILs on pCR
To further assess the predictive effect of TIL detection in breast cancer patients with different molecular subtypes, the OR value of pCR was analysed by meta- REC Representativeness of the exposed cohort, SNEC Selection of the non exposed cohort, AE Ascertainment of exposure, DO Demonstration that outcome of interest was not present at start of study, SC study controls for age, sex, AF study controls for any additional factor, AO Assessment of outcome, FU follow-up long enough for outcomes to occur (36 Months), AFU Adequacy of follow up of cohorts (≥90%)."1" means that the study is meeted the item and "0" means the opposite situation analysis. In this meta-analysis, we chose studies that focused on TILs as a continuous parameter (per 10% increments  Fig. 2 summarizes the results of the pCR assessment. Publication bias was detected by the funnel plot (Fig. 5a). Egger's test indicated that there was publication bias.  Fig. 3 summarizes the results of the DFS assessment. Publication bias was detected by the funnel plot. No significant publication bias was found (Fig. 5b) . High-quality studies (NOS score > 6) were used to conduct the sensitivity analysis, and the results were consistent. Fig. 4 summarizes the results of the OS assessment. Publication bias was tested by the funnel plot. No significant publication bias was found (Fig. 5c). Egger's test indicated that there was not publication bias.

Discussion
Breast cancer is a highly heterogeneous disease in terms of its clinical processes and molecular types. At present, standardized systemic therapy has significantly increased the survival of breast cancer patients, but metastasis and recurrence remain the determinants of death. Therefore, Fig. 3 Forest plot of HR for DFS. Pooled assessing HR for DFS how to further reduce recurrence and metastasis is still a key issue in clinical practice. The complex interaction between the immune system and cancer cells plays a vital role in controlling and eradicating cancer [3]. A few decades ago, people noticed that the tumour microenvironment contained a variable number of lymphocytes, later called tumour-infiltrating lymphocytes or TILs [43]. TILs have become a potential biomarker for survival prediction in breast cancer patients [4,5]. In patients with different molecular subtypes, a comprehensive evaluation of the clinical impact of TILs will help to uncover the important mechanisms of the interaction between tumour and host immunity. Nevertheless, the clinical significance of TILs in patients with different molecular subtypes is still unclear. By summarizing and analysing relevant high-quality studies, our meta-analysis aims to provide evidence for determining the clinical significance of TILs in the different molecular subtypes of breast cancer.
The pooled analysis confirmed that LPBC was significantly correlated with higher histopathological grade. Moreover, our meta-analysis indicated that LPBC was related to Ki-67, ER and PR status. Afterwards, sensitivity analysis excluding low-quality studies showed consistent results. Whether TILs play different roles in patients with different molecular subtypes remains unknown. We further analysed the prognostic value and predictive roles of TILs in patients with different molecular subtypes. To further estimate the survival impact of TIL detection in patients with different molecular subtypes, the HR values of DFS and OS were analysed by meta-analysis. The assessed pooled OR value confirmed that high TILs were correlated with significantly increased pCR rates in patients with the HER2-enriched   For the luminal molecular subtype of breast cancer, there was no significant increase in the pCR rate in the high TIL group. In addition, there was no significant improvement in DFS in the high TIL group. Conversely, the pooled analysis confirmed that the high TIL group of the luminal molecular subtype of breast cancer was significantly correlated with unfavourable OS. Considering the small number of studies, the results of this analysis should be interpreted with caution.
Our meta-analysis confirmed that TILs are an ideal biomarker for TNBC and the HER2-enriched molecular subtype of breast cancer in the prediction of pCR and favourable prognosis. In contrast, TILs are a biomarker for predicting poor OS in the luminal molecular subtype of breast cancer. Therefore, TILs should be monitored in breast cancer patients for rational stratification and adjustment of the treatment strategy, and further detailed and in-depth studies on TILs and breast cancers of different molecular subtypes are needed. Further study on the different roles of different TIL subclasses in the different molecular subtypes of breast cancer will help us further understand the precise mechanisms of TILs and provide more evidence for the immunotherapy of breast cancer with different molecular subtypes.
The limitations of this meta-analysis include the following aspects. First, heterogeneity cannot be avoided completely, so we chose a random effect model. Second, fewer high-quality stratified studies on the different molecular subtypes of breast cancer can affect the statistical efficacy of our results. Therefore, it is necessary to conduct more prospective clinical studies to clarify the true usefulness of TILs. Third, our study was based on data provided by different studies, not individual patient data, so reliable correlation estimates could not be made. Although our research has some limitations, we systematically evaluated a large number of high-quality studies with multivariate analysis, and the research results may be a reliable reference for guiding clinical practice.

Conclusions
In conclusion, we performed a meta-analysis including thirty-three high-quality studies that implemented multivariate analysis, and 18,170 patients with different molecular subtypes of breast cancer were analysed. Our meta-analysis confirms that high TILs are correlated with favourable survival and predict pCR in breast cancer patients with TNBC and the HER2-enriched molecular subtype. Conversely, the pooled analysis confirmed that the high TIL group of the luminal molecular subtype of breast cancer was significantly correlated with unfavourable OS. Large-scale, multicentre and welldesigned high-quality studies are needed to study the role of different TIL subclasses in the different molecular subtypes of breast cancer. Moreover, it can provide guidance for the clinical practice of breast cancer with different molecular subtypes.